diffusers 0.29.2__py3-none-any.whl → 0.30.1__py3-none-any.whl

diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py

@@ -22,12 +22,10 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention_processor import (
     AttnProcessor2_0,
-    LoRAAttnProcessor2_0,
-    LoRAXFormersAttnProcessor,
     XFormersAttnProcessor,
 )
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -68,7 +66,11 @@ def preprocess(image):
 
 
 class StableDiffusionUpscalePipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin, FromSingleFileMixin
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
+    FromSingleFileMixin,
 ):
     r"""
     Pipeline for text-guided image super-resolution using Stable Diffusion 2.
@@ -78,8 +80,8 @@ class StableDiffusionUpscalePipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
-        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
 
     Args:
@@ -245,7 +247,7 @@ class StableDiffusionUpscalePipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -378,7 +380,7 @@ class StableDiffusionUpscalePipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 
@@ -520,8 +522,6 @@ class StableDiffusionUpscalePipeline(
             (
                 AttnProcessor2_0,
                 XFormersAttnProcessor,
-                LoRAXFormersAttnProcessor,
-                LoRAAttnProcessor2_0,
             ),
         )
         # if xformers or torch_2_0 is used attention block does not need
@@ -616,7 +616,7 @@ class StableDiffusionUpscalePipeline(
         >>> # load model and scheduler
         >>> model_id = "stabilityai/stable-diffusion-x4-upscaler"
         >>> pipeline = StableDiffusionUpscalePipeline.from_pretrained(
-        ...     model_id, revision="fp16", torch_dtype=torch.float16
+        ...     model_id, variant="fp16", torch_dtype=torch.float16
         ... )
         >>> pipeline = pipeline.to("cuda")
 

diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py

@@ -20,7 +20,7 @@ from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokeniz
 from transformers.models.clip.modeling_clip import CLIPTextModelOutput
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, PriorTransformer, UNet2DConditionModel
 from ...models.embeddings import get_timestep_embedding
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -58,7 +58,9 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin):
+class StableUnCLIPPipeline(
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
+):
     """
     Pipeline for text-to-image generation using stable unCLIP.
 
@@ -67,8 +69,8 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
-        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         prior_tokenizer ([`CLIPTokenizer`]):
@@ -326,7 +328,7 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -459,7 +461,7 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py

@@ -20,7 +20,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.embeddings import get_timestep_embedding
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -70,7 +70,7 @@ EXAMPLE_DOC_STRING = """
 
 
 class StableUnCLIPImg2ImgPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
 ):
     """
     Pipeline for text-guided image-to-image generation using stable unCLIP.
@@ -80,8 +80,8 @@ class StableUnCLIPImg2ImgPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
-        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         feature_extractor ([`CLIPImageProcessor`]):
@@ -290,7 +290,7 @@ class StableUnCLIPImg2ImgPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -423,7 +423,7 @@ class StableUnCLIPImg2ImgPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

diffusers/pipelines/stable_diffusion_3/__init__.py

@@ -25,6 +25,7 @@ except OptionalDependencyNotAvailable:
 else:
     _import_structure["pipeline_stable_diffusion_3"] = ["StableDiffusion3Pipeline"]
     _import_structure["pipeline_stable_diffusion_3_img2img"] = ["StableDiffusion3Img2ImgPipeline"]
+    _import_structure["pipeline_stable_diffusion_3_inpaint"] = ["StableDiffusion3InpaintPipeline"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
@@ -35,6 +36,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     else:
         from .pipeline_stable_diffusion_3 import StableDiffusion3Pipeline
         from .pipeline_stable_diffusion_3_img2img import StableDiffusion3Img2ImgPipeline
+        from .pipeline_stable_diffusion_3_inpaint import StableDiffusion3InpaintPipeline
 
 else:
     import sys

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -29,9 +29,12 @@ from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
 from ...utils import (
+    USE_PEFT_BACKEND,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
@@ -329,6 +332,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         clip_skip: Optional[int] = None,
         max_sequence_length: int = 256,
+        lora_scale: Optional[float] = None,
     ):
         r"""
 
@@ -374,9 +378,22 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_skip (`int`, *optional*):
                 Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                 the output of the pre-final layer will be used for computing the prompt embeddings.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
         """
         device = device or self._execution_device
 
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
         prompt = [prompt] if isinstance(prompt, str) else prompt
         if prompt is not None:
             batch_size = len(prompt)
@@ -479,6 +496,16 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
                 [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1
             )
 
+        if self.text_encoder is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     def check_inputs(
@@ -683,7 +710,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
                 Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
                 in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
                 passed will be used. Must be in descending order.
-            guidance_scale (`float`, *optional*, defaults to 5.0):
+            guidance_scale (`float`, *optional*, defaults to 7.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
@@ -746,8 +773,8 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         Examples:
 
         Returns:
-            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] or `tuple`:
-            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] if `return_dict` is True, otherwise a
             `tuple`. When returning a tuple, the first element is a list with the generated images.
         """
 
@@ -787,6 +814,9 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
 
         device = self._execution_device
 
+        lora_scale = (
+            self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+        )
         (
             prompt_embeds,
             negative_prompt_embeds,
@@ -808,6 +838,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_skip=self.clip_skip,
             num_images_per_prompt=num_images_per_prompt,
             max_sequence_length=max_sequence_length,
+            lora_scale=lora_scale,
         )
 
         if self.do_classifier_free_guidance:

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

@@ -25,13 +25,17 @@ from transformers import (
 )
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import SD3LoraLoaderMixin
 from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
 from ...utils import (
+    USE_PEFT_BACKEND,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
@@ -62,7 +66,7 @@ EXAMPLE_DOC_STRING = """
         >>> pipe = pipe.to(device)
 
         >>> url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
-        >>> init_image = load_image(url).resize((512, 512))
+        >>> init_image = load_image(url).resize((1024, 1024))
 
         >>> prompt = "cat wizard, gandalf, lord of the rings, detailed, fantasy, cute, adorable, Pixar, Disney, 8k"
 
@@ -346,6 +350,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         clip_skip: Optional[int] = None,
         max_sequence_length: int = 256,
+        lora_scale: Optional[float] = None,
     ):
         r"""
 
@@ -391,9 +396,22 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
             clip_skip (`int`, *optional*):
                 Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                 the output of the pre-final layer will be used for computing the prompt embeddings.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
         """
         device = device or self._execution_device
 
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
         prompt = [prompt] if isinstance(prompt, str) else prompt
         if prompt is not None:
             batch_size = len(prompt)
@@ -496,6 +514,16 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
                 [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1
             )
 
+        if self.text_encoder is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     def check_inputs(
@@ -605,8 +633,6 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
             )
 
         image = image.to(device=device, dtype=dtype)
-        if image.shape[1] == self.vae.config.latent_channels:
-            init_latents = image
 
         batch_size = batch_size * num_images_per_prompt
         if image.shape[1] == self.vae.config.latent_channels:
@@ -726,7 +752,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
                 Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
                 in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
                 passed will be used. Must be in descending order.
-            guidance_scale (`float`, *optional*, defaults to 5.0):
+            guidance_scale (`float`, *optional*, defaults to 7.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
@@ -785,8 +811,8 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
         Examples:
 
         Returns:
-            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] or `tuple`:
-            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] if `return_dict` is True, otherwise a
             `tuple`. When returning a tuple, the first element is a list with the generated images.
         """
 
@@ -854,7 +880,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
         # 4. Prepare timesteps
         timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
         timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
-        latent_timestep = timesteps[:1].repeat(batch_size * num_inference_steps)
+        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
 
         # 5. Prepare latent variables
         if latents is None: